ThursdAI - The top AI news from the past week · From Weights & Biases, Join AI Evangelist Alex Volkov and a panel of experts to cover everything important that happened in the world of AI from the past week

ThursdAI - Mar 5 - OpenAI's GPT-5.4 Solves a 20-Year Math Problem, Anthropic Gets Designated a Supply Chain Risk, Qwen Drama Unfolds

·1 hr 36 min·5 clips

Bartosz Naskerski says GPT 5.4 solved a 20-year research math problem.

OpenAI drops GPT-5.4 Thinking and GPT-5.4 Pro. The show frames it as the first generalized model since GPT-5.2. It also folds the coding capabilities of GPT-5.3 Codex into a unified reasoning model. That alone would be enough for a busy week, but the episode does not settle down there. The crew tests the model live on air, and the rough edges show up in real time, which makes the segment feel less like a product recap and more like a stress test with microphones on. A huge beat lands when Bartosz Naskerski says 5.4 solved a research-level frontier math problem he had worked on for about 20 years. He calls it his personal move 37. The reaction is immediate, because the solution is described as very clean, nice, and almost human. From there the conversation turns to where speed actually matters. Low-latency work comes up first. Logistics and robotic experiments are the obvious examples. The panel also notes that some situations may still favor something like Codex Spark if the price and performance balance works better. That keeps the episode anchored in tradeoffs instead of pure wow factor. The show then reminds listeners that this version of ChatGPT is the free default for many users. That matters because the free tier is where most of the world still lives, while only a small percentage are paying and able to switch. Google's Gemini Flash Lite arrives next, described as the fastest and most cost-efficient model in Gemini. The comparison set includes Haiku, GPT-5 Mini, and Grok fast, plus a 1 million token context window and roughly 300 tokens per second. The episode closes on the show's newsletter, its three-year run, and the sense that the AI news cycle is moving fast enough to make everyone sound slightly unsteady.

As heard by us

A brisk AI-news roundup that lets the live testing do the talking.

PlayNext's episode opens on OpenAI's new GPT 5.4 Thinking and 5.4 Pro releases, then follows the live testing that makes the model's stumbles and recoveries easy to hear.

Read the full review in PlayNext →

Why you'd press play

Catch the week’s AI chaos before it hardens into lore, product names, and hype.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from ThursdAI - The top AI news from the past week